|
Abstract:
|
It is argued that taking on the challenge of developing a product (a menu-driven software system with user interface for client self-use) unlike a tool (requiring statistical skills to operate) is mutually beneficial for both researchers (for a disciplined R&D, setting priorities, and client input and validation at early stages) and clients (for understanding and confidence about the product and to plan for data gaps and customer needs in advance). We describe some problems and possible solutions for a SAE-product that enhances the existing methods by allowing for smoothing of the error covariance, collapsing of areas with little or no samples, multivariate modeling with combined longitudinal and cross-sectional data, approximate self-benchmarking to estimates of subgroups of areas or target variables, and Kalman filtering for computational efficiency and innovations for diagnostics.
|
